search

jasperzhong / read-papers-and-code

My paper/code reading notes in Chinese
43 stars 3 forks source link

ASPLOS '23 | SparseTIR: Composable Abstractions for Sparse Compilation in Deep Learning #365

Closed jasperzhong closed 11 months ago

jasperzhong commented 11 months ago

https://dl.acm.org/doi/pdf/10.1145/3582016.3582047

jasperzhong commented 11 months ago

SparseTIR，总体来看是想做一个自动生成并优化sparse operator算子的事情，非常general，workload不止包括GNN (SpMM, SDDMM), HGNN (RGCN)，还包括Transformer (sparse attention), 3D点云的sparse cnovolution. design space是包括各种sparse matrix format和多种hardwares. 具体design水平有限看不懂. 实验对于HGNN这一块是跑了RGCN，推理速度相比之前SOTA Graphiler提高了4.2-40.2x.